A High-Throughput MAC Strategy for Next-Generation WLANs Seongkwan Kim Youngsoo Kim Sunghyun Choi Multimedia & Wireless Networking Laboratory Seoul National University {skim,yskim}@mwnl.snu.ac.kr,
[email protected] Abstract Today, IEEE 802.11 Wireless LAN (WLAN) has emerged as a prevailing technology for the broadband wireless networking. Along with many emerging applications and services over WLANs, the demands for faster and highercapacity WLANs have been growing fast. In this paper, we propose a new medium access control (MAC) scheme for the next-generation high-speed WLANs such as IEEE 802.11n. The proposed MAC, called Multi-user polling Controlled Channel Access (MCCA), is composed of two sub-schemes. The first one is multi-user polling in order to achieve higher network utilization. We also introduce a frame aggregation scheme as the another proposed scheme, which performs aggregations at both MAC and physical (PHY) layers, and can achieve even higher throughput gain as a result. From simulations, we confirm that the proposed MCCA scheme enhances the aggregate throughput of non-qualityof-service (non-QoS) traffic by an order of magnitude from 17.4 Mbps to 129.9 Mbps, while the aggregate throughput and QoS requirements continue to be satisfied.
1. Introduction In recent years, IEEE 802.11 Wireless LAN (WLAN) has gained a prevailing position in the market for the (indoor) broadband wireless access networking. The IEEE 802.11 standard defines the medium access control (MAC) layer and the physical (PHY) layer specifications [1]. Due to the contention-based channel access nature of the mandatory channel access function (i.e., distributed coordination function or DCF), it supports only the best-effort service without guaranteeing any quality-of-service (QoS). The new MAC protocol of the 802.11e is called the Hybrid Coordination Function (HCF), which will provide QoS [3, 8, 9, 10]. The standardization of the IEEE 802.11e is still on-going at the final stage. The HCF contains a contentionbased channel access mechanism (referred to as Enhanced Distributed Channel Access or EDCA) and a centralized
Kyunghun Jang Jin-Bong Chang Communication & Network Laboratory Samsung Advanced Institute of Technology {khjang,jinb.chang}@samsung.com
channel access mechanism, namely, HCF Controlled Channel Access (HCCA). Recently, the needs for real-time services, such as Voice over IP (VoIP) over WLANs have been increasing drastically. Due to such fast-growing demands, the market, where IEEE 802.11b has gained its wide deployment base so far, is switching to new IEEE 802.11g-based products supporting up to 54 Mbps [4]. This kind of trend, i.e., growing demands for higher-speed WLANs, will be accelerated further along with the evolution of the wired counterparts, e.g., 1 or 10 Gbit Ethernets. IEEE 802.11 Task Group N (TGn) has been initiated in September 2003 for a high-speed WLAN to provide a maximum throughput of at least 100 Mbps measured at the MAC data service access point (SAP) [7]. This group will generate a high-speed WLAN standard called IEEE 802.11n in the future. Currently, many different technical solutions covering both PHY and MAC are being proposed for the new protocol architecture within TGn [6]. The 802.11 is known to have a high overhead for the MAC/PHY operations such as PHY preamble/header, MAC header, acknowledgment, and backoff procedures, thus yielding the throughput performance much worse than the underlying PHY transmission rate. In [11], Yang and Jon demonstrate that by simply increasing the PHY rate without reducing the MAC/PHY overhead, the enhanced throughput is bounded under 100 Mbps, even if the PHY rate goes to infinity. It means that we need to enhance the 802.11 MAC by reducing overheads in order to create the next-generation highspeed WLAN. In this paper, we propose a new MAC scheme, called Multi-user polling Controlled Channel Access (MCCA), which is composed of two schemes including multi-user polling and two-level frame aggregation. The proposed scheme reduces the protocol overheads by eliminating the backoff process via an efficient multi-user polling and reducing the MAC/PHY header overheads via frame aggregation at both MAC and PHY. From simulations, we confirm that the proposed MCCA achieves our design goal by significantly improving throughput performance.
2. Related Work Under the 802.11 PCF or the 802.11e HCCA, each station transmits frames upon being polled. Apparently, pollframe transmissions can be another type of overhead. The number of poll-frames can be reduced by using multi-user polling, and it can enhance wireless channel utilization. Shou-Chih Lo et al. proposed a multi-user polling mechanism for WLAN [12]. Their approach is mainly focused on supporting QoS traffic in order to satisfy the QoS requirement under the DCF environment. However, if one uses the emerging 802.11e EDCA, the QoS traffic can be supported more effectively. In high-speed environment, the effect of overheads mentioned Section 1 increases since these overheads are fixed irrespective of the transmission rate. According to [13], high throughput performance can be achieved via frame aggregation, though the aggregation is performed at the device driver level. However, only the frames with the same destination and the same kind of traffic can be aggregated in [13]. It is surely desired to apply the aggregation of frames with different destination addresses, i.e., downlink frames transmitted from access point (AP) to stations. As it will be mentioned in Section 3, the AP’s downlink access is very important since its performance could be a bottleneck of the entire network performance. In order to achieve highly efficient aggregate throughput, the aggregation of frames with different destinations should be supported.
3. Baseline MAC Protocols In this section, we briefly describe how 802.11e EDCA works since the proposed MCCA works on top of EDCA1 . The IEEE 802.11e EDCA is designed to provide differentiated, distributed channel accesses for frames with 8 different user priorities by enhancing the DCF [3]. Each frame from the higher layer arrives at the MAC along with a specific priority value (ranging from 0 to 7). Then, each QoS data frame carries its priority value in the QoS Control field of the MAC frame header. An 802.11e STA shall implement four channel access functions, where a channel access function is an enhanced variant of the DCF. Each frame arriving at the MAC with a user priority is mapped into an access category (AC). Basically, a channel access function uses AIFS2 [AC], CWmin[AC], and CWmax[AC] instead of DIFS, CWmin, and CWmax, of the DCF, respectively, for the contention process to transmit a frame belong to access category AC. AIFS[AC] is determined by AIF S[AC] = SIF S + AIF SN [AC] · SlotT ime
where AIFSN[AC] is an integer greater than 1 for STAs and an integer greater than 0 for access points (APs). The backoff counter is selected from [0, CW[AC]]. The values of AIFSN[AC], CWmin[AC], and CWmax[AC], which are referred to as the EDCA parameter set, are advertised by the AP via Beacons and Probe Response frames. The AP can adapt these parameter dynamically depending on the network condition. Basically, the smaller AIFSN[AC] and CWmin[AC], the shorter the channel access delay for the corresponding priority, and hence the more capacity share for a given traffic condition. However, the collision probability increases when operating with smaller CWmin[AC]. These parameters can be used in order to differentiate the channel access among different priority traffic. IEEE 802.11e EDCA also provides a new channel access method called Transmission Opportunity (TXOP), which is an interval of time when a particular STA has the right to initiate frame exchange sequences onto the wireless medium. A TXOP is defined by a starting time and a maximum duration. The TXOP is either obtained by the STA by successfully contending or is assigned by the AP. It should be also noted that the AP can use the EDCA parameter values different from the announced ones for the same AC. The 802.11 DCF originally is designed to provide a fair channel access to every station including the AP. However, since typically there is more downlink (i.e., AP-to-stations) traffic than uplink (i.e., stations-toAP), AP’s downlink access has been known to be the bottleneck to the entire network performance. Accordingly, EDCA, which allows the differentiation between uplink and downlink channel accesses, can be very useful to control the network performance. The traffic identifier (TID) of a frame is a label, which specifies the corresponding QoS requirements. There are 16 possible TID values, where the values from 0 to 7 specify the user priority value of a frame, and the values from 8 to 15 specify the traffic stream which the frame belongs to. The TID value is specified in the QoS Control field (from bit 0 to bit 3) of the 802.11e QoS data frame’s MAC header. In Section 5.1, we use TID value in the proposed MCCA in order to aggregate frames with different TIDs. In order to improve the MAC efficiency, a new acknowledgment scheme, called Block Ack (BA), is defined in IEEE 802.11e. The scheme basically works as follows: during a TXOP, a STA (or AP) can transmit a number of frames without receiving any Ack. After all the frame transmissions within the TXOP are completed, the STA (or AP) sends a control frame, called Block Ack Request (BAR). The receiving STA (or AP) of the BAR should respond with BA 3 .
1 We
assume the 802.11a PHY in this paper because the target of our scheme is the next-generation high-speed WLAN such as TGn. 2 AIFS : Arbitration Interframe Space [3]
3 There are two types of Block Ack procedures in 802.11e, which are Immediate and Delayed Block Ack. The former is explained here.
MAC Protocol Data Unit (MPDU) F MAC Header MSDU C S DA1:TID1
F MAC Header MSDU C S DA1:TID1
F MAC Header MSDU C S DA1:TID2
DATA MPDU
M D
M CHDATA - 1 M CHDATA - 2 D D MPDU MPDU
PSDU
P D
rate 1 basic rate
2
6
0 ~ 2304
4
Sequence Control
Address 4
Frame Body
FCS
DATA MPDU
Aggregate PHY Protocol Data Unit (PPDU) P Preamble/ PLCP Header D
6 Address 3
Figure 2. Type-1 compressed header data
Aggregate PHY Service Data Unit (PSDU) M D
2 Frame Control
Octets: F MAC Header MSDU C S DA2:TID1
PSDU
Octets:
2
6
2
6
2
0 ~ 2304
4
Frame Control
Address 3
Sequence Control
Address 4
QoS Control
Frame Body
FCS
rate 2
Figure 3. Type-2 compressed header data Figure 1. Two-level frame aggregation
4. Underlying PHY Model The purpose of this paper is to enhance aggregate throughput and network channel utilization. As mentioned in Section 1, the overheads such as PHY preamble/header, MAC header, ACK, and backoff procedure should be reduced in order to achieve the goal. Moreover, we also consider faster PHY technology to achieve more than 100 Mbps throughput gain at the MAC SAP which is one of the goals of the 802.11 TGn. Accordingly, we assume Multiple Input Multiple Output (MIMO) technology as the underlying PHY model. MIMO is one of the underlying PHY schemes being considered for 802.11 TGn. However, MIMO is considered as the dominant technology of the next-generation WLAN in the nearest time4 . In this paper, we assume that the PHY model uses 2 × 2 MIMO, which can provide 2 times faster transmission rate than 802.11a PHY. In addition, we assume that our PHY in consideration uses multiple channel bonding scheme. That is, two 802.11a channels (of 20 MHz) are combined together, and the bonded channel of 40 MHz is used for the communications. Therefore, we can achieve 2 times faster rate due to this bonding scheme. Accordingly, if we employ both 2 × 2 MIMO and two channel bonding schemes, the PHY transmission rate, which is 4 times faster than that of 802.11a PHY (i.e., up to 216Mbps), can be achieved.
5. Multi-user polling Controlled Channel Access (MCCA) In this paper, we propose two MAC schemes to enhance network utilization and maximize aggregate throughput, namely, two-level frame aggregation and multi-user polling. The name of MCCA, i.e., Multi-user polling Con4 At the last 802.11 TGn meeting on Sep. 2004, while there are 36 proposals offered as how 802.11n will look eventually, all the proposals use MIMO.
trolled Channel Access, is coined from the access manner to the wireless channel of our scheme.
5.1. Two-Level Aggregation As part of MCCA, we employ a two-level frame aggregation scheme, which is composed of two types of hierarchical aggregations, namely, aggregate PHY Service Data Unit (PSDU) and aggregate PHY Protocol Data Unit (PPDU) schemes. Proposed aggregation schemes are performed at both MAC and PHY layers, respectively, according to the characteristics of the aggregated frames, such as the destination address and TID. Fig. 1 illustrates how the proposed two-level aggregation works. Multiple MPDUs addressed to the same STA are aggregated into an aggregate PSDU. Each aggregated MPDU follows a MPDU Delimiter (MD), which plays a role to robustly delimit the MPDUs within the aggregate PSDU. An MD is composed of several fields including Unique pattern, CRC, and MPDU length. The aggregate PSDU is for a single destination. The first case of the aggregate PSDU scheme is as follows. If there are buffered data MPDUs with the same destined address and the same TID, they can be aggregated into a PSDU. The first data MPDU has the legacy MAC header, and the subsequent MPDUs use the format of CHDATA-1, i.e., Type1 Compressed Header DATA, and the format of this new frame is illustrated in Fig. 2. Because all the MPDUs are separated by MD, the receiving high throughput STA (HT STA)5 can recognize the existence of another MPDU after the MD. The CHDATA-1 header should be rebuilt based on the first MPDU’s MAC header by the receiver. If there are buffered data MPDUs with the same destination address, but different TIDs, they can also be aggregated into a PSDU, and this type of aggregation is the second case of the aggregate PSDU. The rule to make an aggregate PSDU is the same as the first case mentioned 5 We refer a STA, which supports the MCCA proposed in this paper, to as an HT STA.
above, except for the compressed header format, i.e., Type2 Compressed Header DATA (CHDATA-2) is used instead of legacy MAC header for the subsequent MPDUs. The format of CHDATA-2 header is shown in Fig. 3. In this CHDATA-2 header, QoS field is needed since MSDUs contained in this MPDU have different TIDs. Similar to the operation for receiver to rebuild the CHDATA-1 header, the CHDATA-2 header should be rebuilt on the equivalent way by receiver. For the composite case of the above two cases, i.e., there are buffered data MPDUs with the same and different TIDs, MPDUs which have either CHDATA-1 or CHDATA2 headers can be aggregated into a PSDU. In this case, the first MPDU should have the legacy MAC header and it should be referenced for rebuilding the MAC header of CHDATA-1 and CHDATA-2 MPDUs. Fig. 1 is an example of this case of aggregate PSDU. Apparently, with the header compression scheme, we can use channel bandwidth more efficiently and transmit more data during a fixed channel access opportunity. The second-level aggregation, referred to an aggregate PPDU, can be applied if there are buffered PSDUs with different destination addresses. With the aggregate PPDU scheme, the buffered PSDUs can be aggregated into a PPDU without additional preambles as shown in Fig. 1. If a number of PSDUs are aggregated into a PPDU, PSDU Delimiters (PDs) with own unique patterns precede each aggregate PSDU, and this allows to transmit each PSDU at a different rate according to the optimal rate of the destined HT STA. A PD should be transmitted at the basic rate, i.e., 6 Mbps of the 802.11a PHY, in order for all the HT STAs to be able to decode it. There is a limit of aggregate PPDU. According to the 802.11a, the PLCP in a PPDU has Length (of 12 bits) and Rate (of 4 bits) fields in the legacy SIGNAL field. The Length and Rate fields are virtually set in order to cover the duration of the PPDU. The length of an aggregate PPDU is limited so that its duration may not exceed 5.46 milliseconds, because the maximum duration of the PPDU in 802.11a is 5.46 millisecond, which can be derived from the maximum length (4095 bytes) divided by the lowest rate (6 Mbps). In other words, an aggregate PPDU can spoof only the duration up to 5.46 milliseconds. Basically, we assume the infrastructure mode of the IEEE 802.11 because a kind of controlled channel access scheme is proposed in our scheme. Accordingly, there are two different directions of data transmission, i.e., uplink and downlink and the policies, which the proposed two-level aggregation scheme is applied, are different depending on the transmission direction. In an uplink phase (U/L), in which uplink data is transmitted, an HT STA gaining access to the channel can transmit a PPDU containing an aggregate PSDU destined to HT AP. In this phase, HT STAs
octets : 2
Frame Control
2
6
Dur/ID
BSSID
2
2
D/L U/L Count(n) Count(m)
5*n
4*m
4
D/L MAP
U/L MAP
FCS
Figure 4. Format of MP-frame
may transmit more than one aggregate PSDUs after SIFS if the transmission is finished within a given TXOP limit. However, the aggregate PPDU scheme cannot be used for this phase because the destination of a HT STA is always the AP. On the other hand, HT AP can utilize the aggregate PPDU scheme as well as the aggregate PSDU for the downlink phase (D/L) when the controlled access scheme is used, which will be explained in detail in the next subsection.
5.2. Multi-user Polling As the second part of MCCA, we propose a controlled channel access scheme, based on multi-user polling. Our multi-user polling scheme, which we referred to as multipolling, has the advantages of high channel utilization. One of the main features of our multi-user polling is that, unlike PCF and HCCA, it can deal with downlink traffic streams during the controlled access period. Moreover, a modified hybrid coordinator (HC) 6 in our scheme can send multiple polls at the same time. To support these new and intelligent features, the format of the multiple poll frame, namely, MP-frame, should be designed well. Fig. 4 shows the format of MP-frame. The MP-frame provides the control of flows for D/L and U/L within a service period. It conveys a number of D/L MAPs and the number of U/L MAPs, where the MAPs handle the aggregation exchanges during a service period. The Dur/ID field is used to set a long NAV value to protect the service period as soon as the receipt of the MPframe at the legacy STAs associated to the HT AP. The D/L Count field and U/L Count field, respectively, indicate the number of PSDUs in D/L to be used by the HT AP and the number of HT STAs to have a transmission opportunity in U/L. They shall be referenced by D/L MAP and U/L MAP fields which are successively repeated within the MP-frame. Fig. 5 illustrates how the overall access mechanism of multipolling works. Whenever an HT AP transmits an MPframe, it initiates a service period. The HT AP gains the access to the channel by transmitting an MP-frame after waiting for a PCF Interframe Space (PIFS) idle time interval, which is shorter than DIFS and any AIFS. When the STA receives the MP-frame, it sets the backoff counter to the appropriate value, that is implicitly assigned by AP in the MP6 HC
is the new notion in the IEEE 802.11e specification [3].
HT STA1 TXOP Limit
0
Preamble PLCP
Preamble PLCP MBA MPDU Preamble PLCP MBA MPDU
Aggregate PSDU
Preamble PLCP MBA MPDU Data MPDU Data MPDU MBAR MPDU
HT AP
HT STA2
2
1 0
Aggregate PSDU
1
0
Preamble PLCP Data MPDU MBAR MPDU
HT STA1
1
Preamble PLCP MBA MPDU Data MPDU Data MPDU MBAR MPDU
0
HT STA2
HT STA3
TXOP Limit
4 3
PIFS
DIFS
SIFS
SIFS
DIFS
SIFS
SIFS
SIFS
PIFS
Figure 6. Example of MCCA Block Ack
Service Period PIFS
DIFS
2 1 0 DATA
NAV
Non-polled (HT) STA4 PIFS
CF-END Frame
Within TXOP limit
Preamble PLCP MBA MPDU Data MPDU MBAR MPDU
Aggregate PPDU
Preamble PLCP Data MPDU Data MPDU MBAR MPDU Data MPDU Data MPDU MBAR MPDU
HT AP
Preamble PLCP MP-Frame
MP
Within TXOP limit
CFEND
Aggregate Aggregate PSDU PSDU
Preamble PLCP MBA MPDU
collision-free CSMA/CA
dynamic TDM
DIFS
PIFS
AIFS[AC]
Figure 5. Multi-user polling operation
frame according to the polling order. It is also possible that CF-End frame may be used when there is no more uplink traffic from the polled HT STAs, which this functionality is the same as that of PCF. During a service period, AIFS is set to DIFS irrespective of the AC of the frames being transmitted. For the D/L, which starts immediately after the MPframe transmission with a PIFS deference, the aggregate PPDU scheme is applied, and hence we referred to this D/L as dynamic Time Division Multiplexing (TDM), because each HT STA receives a PSDU destined to itself at its receiving time. After the end of the D/L, uplink transmission opportunity is handed over to the HT STA, whose backoff counter becomes zero. In this U/L, each HT STA follows EDCA rule except that its backoff number is assigned from HC in advance with AIFS value set to DIFS in all occasions. Within a TXOP of each HT STA scheduled via the U/L MAP, an HT STA can transmit multiple frames including one or more aggregate PSDUs consecutively. If an HT STA scheduled via the U/L MAP does not transmit any frame, the next scheduled HT STA can transmit after a backoff slot time according to the EDCA channel access rule. An HC shall know the decrease of the backoff counters while the STAs transmit their data according to the U/L Count field in the MP-frame. The backoff counter of the last STA decreases from U/L Count value to zero eventually at the end of the service period. If the uplink transmission finishes and the service period remains enough to transmit the CF-End frame, then the HC sends the CF-End frame to inform both HT STAs and legacy STAs of the end of the multipoll operation before the originally-scheduled end of the service period. In a controlled channel access scheme such as PCF and MCCA, the whole network operation can be collapsed by hidden stations. The hidden station problem has been a inveterate obstacle in order for these centralized access
schemes to operate smoothly until now. The emerging IEEE 802.11k [5], however, defines hidden node report method using a hidden station detection scheme, which basically works as follows. First of all, if STA i detects a frame, which the AP transmits to STA j via downlink, but STA i cannot detect the corresponding Ack from STA j. If this happens persistently,7 and STA i never receives any frame transmitted by STA j, STA i can conclude that STA j is hidden from itself. By using this mechanism, AP can manage the STA list, in which the STAs are not hidden each other. Accordingly, if we use 802.11k hidden node report scheme described above in our MCCA scheme, our centralized channel scheme is getting more robust against hidden STAs. From now on, we assume that the proposed MCCA use the hidden node report scheme basically.
5.3. Error Recovery via Block Ack There can be transmission errors due to the unreliable channel behavior caused by background noise, fading, and so on. As we take into account an aggregation scheme as a major part of the proposed scheme, an Ack mechanism such as Block Ack, which is defined in IEEE 802.11e, is needed. We propose a Block Ack mechanism which is appropriate in our MCCA scheme. We assume that the frame formats of proposed BA and BAR are identical to those defined in 802.11e. However, the usage is different from the 802.11e’s. For example, the proposed BA is a delayed Ack for a number of aggregated transmitted frames, not one for a chain of frames, 802.11e Block Ack. Our BA and BAR themselves can also be aggregated into an aggregate PPDU frame or an aggregate PSDU frame. Therefore, we use the new term of MCCA Block Ack (MBA) and MCCA Block Ack Request (MBAR) to denote them in this paper. Fig. 6 shows an example of error recovery by the proposed MBA. After an MP-frame transmission and PIFS time interval, the HT AP transmits aggregated PSDUs destined to multiple HT STAs including an MBAR to each HT STA. By uplink transmission rule mentioned above, the first 7 Note
that this can happen occasionally due to the channel error.
EDCA
Internet streaming (2Mbps)
Video conferencing (1Mbps)
Throughput (Mbps)
Table 1. Application Characteristics Delay Bound Offered Load Application (msec) (Mbps) VoIP 30 msec 0.096 Video conferencing 100 msec 1 A/V streaming 200 msec 2 Internet file transfer N/A N/A Local file transfer N/A N/A
1
VoIP (0.096Mbps)
0.1
HT STA transmits its own aggregated MPDUs to the HT AP including an MBA MPDUs for the aggregated MPDUs from HT AP and an MBAR MPDU. As shown an example of Fig. 6, if some of MPDUs are not received successfully, the HT STA informs the HT AP of these errors using the Block Ack Bitmap. If there are any error frames as shown in Fig. 6, the HT AP shall retransmit the frame with the corresponding MBAR. There is another example of the transmission error recovery procedure at the right side of Fig. 6. If an HT STA obtaining a channel access opportunity transmits aggregated MPDUs and some errors occur, the HT AP shall inform the HT STA of the erroneous MPDUs by using Block Ack bitmap like the first case error recovery operation as mentioned above.
0
10
20
30 Flow Number
40
50
60
Figure 7. Throughput in Large Enterprise model
In this section, we comparatively evaluate the performance of the IEEE 802.11e EDCA and our MCCA using the ns-2 simulator [14].
which satisfy their corresponding delay bound, are counted to the aggregate throughput calculation. Basically, we assume that all the STAs are fully controlled by the HC in service periods, and hence that there is not any collision which can cause the transmission error. As mentioned in Section 5.3, however, other error sources such as RF interference, channel fading, and so on can exist. To take into account these erroneous conditions, we apply strict error model in which a system can suffer from errors per MPDU-basis at a randomized and uniformly distributed rate. The EDCA parameter set used for each traffic type are the default values defined in [3].
6.1. Simulation Environments
6.2. Simulation Results and Discussion
For the simulation scenario, we use one of the models, called an Large Enterprise scenario, defined by TGn [7]. Three different types of traffic are considered for our simulation, namely, voice, video, and data. The voice traffic is modeled by a two-way constant bit rate (CBR) session according to G.711 codec. The video traffic is modeled by three kinds of applications, including video conferencing and the A/V Internet streaming. The data traffic is modeled by a unidirectional FTP/TCP flow to represent a file transfer. These traffic models are described in Table 1 in detail. In the simulation, we use the 802.11a parameters except the transmission rate. We assume that all the stations use the transmission rate of 216 Mbps, which can be achieved by MIMO-aware PHY techniques described in Section 4. With the Large Enterprise scenario, we compare the throughput performances of (1) EDCA, (2) EDCA with aggregate PSDU, and (3) MCCA. When we consider different types of QoS traffic including voice and video, only the packets
Fig. 7 presents the throughput performance of the EDCA scheme. The first 30 flows are downlink flows and the other 30 flows are uplink flows. Each station has one downlink and one uplink flow, respectively, e.g., flow i and flow 30+i are for station i. A flow with zero throughput represents that there is no active flow. There are three kinds of QoS flows, which are indicated in Fig. 7. The remaining flows are nonQoS flows. In order to show the low throughput values more clearly, e.g., those of VoIP traffic, we use the logarithmic scale for the y-axis. From the simulation results, all of the QoS traffic flows are found to satisfy their QoS requirement with default EDCA parameters. Fig. 8 shows the aggregate throughput performance of all three schemes. The proposed MCCA can achieve very high throughput performance. Approximately, 7.5-time larger aggregate throughput gain can be achieved by MCCA compared to EDCA. This result is even higher than that of aggregate PSDU with 2.2-time gain, approximately.
6. Performance Evaluations
140 EDCA Aggregate PSDU MCCA
EDCA Aggregate PSDU MCCA
129.9
delay bound : 30
100 End-to-end Delay (msec)
Aggregate Throughput (Mbps)
delay bound : 200 delay bound : 100
100
120
80
59.3
60
10
1
40
17.4
20 9.2 0
0.1 QoS traffics
non-QoS traffics
VoIP
Figure 8. Aggregate throughput performance
Downlink AC_BE queue size (packet)
120
100
80
60
40
20
0
AC_BE 17
17.1
17.2
17.3
17.4
17.5
Time (sec)
Figure 9. Downlink EDCA queue status with aggregate PSDU scheme
To evaluate the performance of our new MAC scheme, first of all, we compare aggregate PSDU scheme to EDCA, with the same simulation model as shown in Fig. 7. We apply the aggregate PSDU scheme to all stations including AP, and hence it improves the aggregate throughput of non-QoS traffic by 3.4 times, approximately, while throughput of all QoS traffic still remain the same as those of the EDCA. However, if the characteristic of non-QoS traffic, i.e., TCP flows, is considered, the more improvement can be achievable. It is because the TCP flow has the burst characteristics. It means that more TCP Ack packets are received by the TCP sender, and hence more packets will be sent up to the TCP receive window size. Therefore, we check the status of EDCA queues at the AP to determine whether these EDCA downlink queues are bottleneck or not, in the simulation of the aggregate PSDU scheme. Fig. 9 shows the status of the downlink EDCA queues
Video Conferencing
Streaming A/V
Figure 10. Comparison of end-to-end delay
corresponding to AC BE. Approximately, 120 packets are buffered at the AC BE queue almost always. We use TCP new Reno with Delayed Ack option with TCP receive window size of 6. Since there are 16 downlink and 10 uplink TCP flows, the maximum number of packets, which can be buffered at the AC BE queue, is 126 packets. This means that the downlink AC BE queue is a bottleneck in our simulation scenario when the aggregate PSDU MAC scheme is used. Accordingly, if those downlink packets (AC BE) can be served faster, the aggregate throughput will be increased. This is the reason why we could achieve a very high throughput performance with MCCA as shown in Fig. 8. Even though we get higher throughput gain with MCCA than EDCA scheme or EDCA with aggregate PSDU scheme, we lose nothing. In Fig. 8, this can be confirmed by observing the aggregate throughput of QoS traffic. Fig. 10 also shows that the MCCA does not sacrifice the delay performance of QoS traffic either. As shown in Fig. 10, which has a log-scale y-axis, there are still enough margins to each delay bound for all three schemes. Until now, we have simulated in the environments, where the channel errors do not occur, thus not utilizing the proposed error recovery mechanism described in Section 5.3. Fig. 11 shows the effect of the error recovery mechanism when the frame (MPDU) error rate varies. In this figure, with a log-scale y-axis, the performance of MCCA nonQoS traffic is observed lower than that in Fig. 8 due to additional frame overheads such as MBA and MBAR frames. For this simulation, we assume that the sender retransmits a frame at most once. In other words, even if a frame suffers consecutive errors, it is retransmitted just once. We first observe that both MCCA and EDCA sustain the throughput for QoS traffic irrespective of the frame error rate. On the other hand, the throughput for non-QoS traffic decreases as the frame error rate increases for both systems. However,
References
Throughput (Mbps)
100
10
MCCA : non-QoS MCCA : QoS EDCA : non-QoS EDCA : QoS 1 0
2
4 6 Frame error rate (%)
8
10
Figure 11. Error recovered performance
non-QoS traffic of MCCA still preserves 72% of the overall throughput even if frame errors occur with the rate of 10%, As a result, we can conclude that MCCA with the error recovery mechanism still has a great throughput gain compared to EDCA even in unreliable channel environments.
7. Conclusion In this paper, we propose a new high-efficient MAC scheme, called MCCA, for the next-generation high-speed WLANs. The MCCA is based on the EDCA-based multiuser polling (multipolling) and a two-level frame aggregation scheme performing aggregation at both MAC and PHY layers. Therefore, the proposed scheme can aggregate frames with different QoS requirements and different destinations. The aggregate PPDU scheme, however, has the nature that this aggregated frame should be well scheduled in order for the receiving HT STAs to recognize easily and correctly. Therefore, we have introduced our multipolling for the purpose of scheduling the aggregate PPDU frames. Moreover, by using the multipolling, we also achieve very high efficiency of the channel utilization thanks to using a minimum size of backoff counter value, which is uniquely assigned by U/L MAPs of HT AP. With simulation runs, we show that the MCCA provides very high throughput performance while the delay performance remains acceptable. The achieved performance enhancement is approximately 646 % in terms of the aggregate throughput of non-QoS traffic when the MCCA is employed while the aggregate throughput of QoS traffic is not sacrificed. In the future, we plan to evaluate the proposed scheme further. Those include the consideration of more realistic channel environments such as multipath fading and multiple transmission rates depending on the channel condition of each STA.
[1] IEEE Std. 802.11-1999, Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) specifications, Reference number ISO/IEC 8802-11:1999(E), IEEE Std 802.11, 1999 edition, 1999. [2] IEEE 802.11a-1999, Supplement to Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) specifications: High-speed Physical Layer in the 5 GHz Band, 1999. [3] IEEE 802.11e/D11.0, Draft Supplement to Part 11: Wireless LAN Medium Access Control (MAC) and Physical layer (PHY) specifications: Medium Access Control (MAC) Enhancementsfor Quality of Service (QoS), Oct. 2004. [4] IEEE 802.11g-2003, Supplement to Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) specifications: Further Higher Data Rate Extension in the 2.4 GHz Band, 2003. [5] IEEE 802.11k/D1.0, Draft Supplement to Part 11: Wireless LAN Medium Access Control (MAC) and Physical layer (PHY) specifications: Radio Resource Measurement, July 2004. [6] IEEE 802.11 Working Group Homepage, http://www.ieee802.org/11, online link. [7] Adrian P. Stephens et al., 802.11 TGn Functional Requirements, IEEE 802.11-03/813r12, March 2004. [8] Stefan Mangold, Sunghyun Choi, Guido R. Hiertz, Ole Klein, and Bernhard Walke, “Analysis of IEEE 802.11e for QoS Support in Wireless LANs,” IEEE Wireless Communications Magazine, vol. 10, no. 6, Dec. 2003. [9] Sunghyun Choi, Javier del Prado, Sai Shankar N, and Stefan Mangold, “IEEE 802.11e Contention-Based Channel Access (EDCF) Performance Evaluation,” in Proc. IEEE ICC’03, May 2003. [10] Stefan Mangold, Sunghyun Choi, Peter May, Ole Klein, Guido Hidertz, and Lothar Stibor, “IEEE 802.11e Wireless LAN for Quality of Service,” in Proc. European Wireless (EW’2002), vol. 1, pp. 32-39, Feb. 2002. [11] Yang Xiao and Jon Rosdahl, “Throughput and Delay Limits of IEEE 802.11,” IEEE Communications Letters, vol. 6, no. 8, Aug. 2002. [12] Shou-Chih Lo, Guanling Lee, and Wen-Tsuen Chen, “An Efficient Multipolling Mechanism for IEEE 802.11 Wireless LANs,” IEEE Trans. on Computers, vol. 52, no. 6, June 2003. [13] Youngsoo Kim, Sunghyun Choi, Kyunhun Jang, and Hyosun Hwang, “Throughput Enhancement of IEEE 802.11 WLAN via Frame Aggregation,” in Proc. IEEE VTC’04Fall, Sept. 2004. [14] “The Network Simulator – ns-2,” http://www.isi.edu/nsnam/ns/, online link.
